Re: Linux 3.19-rc5

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Feb 05, 2015 at 10:14:36PM +0100, Bruno Prémont wrote:
> The loop is now replaced by a single WARN() trace - I guess expected:

> From my reading of the thread fixing pccardd/sched TASK_RUNNING usage/check
> is another issue left for the future.

Yeah, something like the below will make it go away -- under the
assumption that that comment is actually correct, I don't know, the
pcmcia people should probably write a better comment :/

Also, set_current_state(TASK_RUNNING) is almost always pointless, nobody
cares about that barrier, so make it go away.

---
 drivers/pcmcia/cs.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/pcmcia/cs.c b/drivers/pcmcia/cs.c
index 5292db69c426..5678e161a17d 100644
--- a/drivers/pcmcia/cs.c
+++ b/drivers/pcmcia/cs.c
@@ -635,6 +635,12 @@ static int pccardd(void *__skt)
 		skt->sysfs_events = 0;
 		spin_unlock_irqrestore(&skt->thread_lock, flags);
 
+		/*
+		 * Supposedly this is a rarely contended mutex and
+		 * sleeping is therefore unlikely, the occasional
+		 * extra loop iteration is harmless.
+		 */
+		sched_annotate_sleep();
 		mutex_lock(&skt->skt_mutex);
 		if (events & SS_DETECT)
 			socket_detect_change(skt);
@@ -679,7 +685,7 @@ static int pccardd(void *__skt)
 		try_to_freeze();
 	}
 	/* make sure we are running before we exit */
-	set_current_state(TASK_RUNNING);
+	__set_current_state(TASK_RUNNING);
 
 	/* shut down socket, if a device is still present */
 	if (skt->state & SOCKET_PRESENT) {